Observe your sacred experiments with mlflow.
Writing experiments with sacred
is great.
mlflow
provides a nice UI that can be used to get a quick overview of your runs and analyze the results.
In your code, add the observer:
from sacred import Experiment
from mlflow_observer import MlflowObserver
from _paths import MY_TRACKING_URI
ex = Experiment('MyExperiment')
ex.observers.append(MlflowObserver(MY_TRACKING_URI))
In the commandline, you can pass a run name through sacred's comment flag:
python train.py -c "My sacred run"
Otherwise the run name will be of the form run_[datetime]
.